Belief in Mean Reversion and the Disposition Effect: An Experimental Test

ABSTRACT

The disposition effect refers to investors' tendency to disproportionately sell more winning than losing assets. The literature mostly offered indirect evidence regarding its cause. Using a lab experiment, the author directly evaluates the two competing behavioral mechanisms for this effect: belief in mean reversion and dual risk attitudes of prospect theory. The participants were endowed with some hypothetical assets, observed price charts, and made selling decisions. Sixty-one percent of them exhibited significant disposition effect. The author elicited the participants' risk attitude parameters and beliefs about price movements, and found that beliefs, especially those in the loss domain, but not the dual risk attitudes, significantly contributed to the between-subject variation of the disposition effect. The results from a goodness-of-fit test also favor the belief mechanism.

KEYWORDS:

• Disposition effect
• Belief in mean reversion
• Prospect theory
• Experiment

Acknowledgments

The author thanks Paul Zak, Joshua Tasoff, Sean Flynn, an anonymous reviewer, participants at the 2013 Bay Area Behavioral and Experimental Economics Workshop, the Western Economics Association International 2013 annual conference, and the Economic Science Association 2013 North-American conference. Southwestern Economic Theory Conference 2014, and seminar participants at Claremont Graduate University and University of Oxford for helpful comments on the manuscript.

Notes

1. For example, among individual investors in the common stock market (Odean [1998]; Grinblatt and Keloharju [2001]), professional futures traders (Heisler [1994]), individual commodity and currency traders (Locke and Mann [2005]), mutual fund managers (Scherbina and Jin [2005]), home sellers (Genesove and Mayer [2001]), and so forth. Experiments, such as that in Weber and Camerer [1998], also successfully replicate behavior consistent with the disposition effect in the lab.

2. Using individual brokerage account data, Odean [1998] showed that the selling of entire winning positions is too aggressive for portfolio rebalancing and that the repurchasing of more losers than winners conflicts with the incentive to avoid high transaction costs. For the examples of failures of rational explanations, see Brown et al. [2006] and Vlcek and Wang [2008].

3. For example, Barberis and Xiong [2009] demonstrated that prospect theory does not generate the disposition effect with its usual parameterization; Hens and Vlcek [2005] showed in an intertemporal choice model that investors would not buy the assets on which they may later exhibit the disposition effect; and Vlcek and Wang [2008] found a lack of correlation between the prospect theory parameters and the disposition effect observed in the lab.

4. Standard economics assume that people use the Bayes' rule, regardless of whether they are buying or selling. However, behavioral theories, such as wishful thinking (Mayraz [2011]), suggest that holding the asset and having a stake may bias beliefs.

5. Rabin and Vayanos [2010] and Barberis, Shleifer, and Vishny [1998] both modeled belief in mean reversion and draw the implication that investors tend to expect short streaks (e.g., the small winners or losers here) to reverse and long streaks to continue.

6. They did not elicit self-reported beliefs, but just used a questionnaire that asks their participants to justify their decision.

7. For instance, in Hung and Yu's [2006] model, disposition investors have standard CRRA utility, but believe in mean reversion.

8. The correlation of the disposition effect with belief in mean reversion and the risk preference parameters can be conveniently proved in the context of this experimental setup. See Appendix 1 for the proof.

9. The lotteries used are in Appendix 2.

10. EC stands for experimental cash, the unit of reward measurement. The exchange rate between EC and U.S. dollars is 1,000 EC = 1 USD.

11. The certainty amount that appears in the first question is the expected value of the lottery. All numbers are rounded to the nearest tens digit.

12. Experimental procedures similar to theirs were replicated and the disposition effect found by Chui [2001] in Macau. Kirchler et al. [2004] provided more experimental evidence for the disposition effect.

13. Without eliciting the participants' beliefs and prospect theory parameters, their test was only indirect. Their evidence against belief in mean reversion in the automatically selling condition rests on the assumption that the automatic liquidation does not alter subjects' belief. However, beliefs may change due to the automatic selling for reasons such as wishful thinking or desirability bias (Mayraz [2011]).

14. The prices took large values so as to be consistent with the order of magnitude in Stage 1 in which risk attitudes were estimated.

15. Instead of calling them upward-drifted and downward-drifted processes, the experiment instructions used neutral words and refered to them as the red and green processes.

16. Another reason for using simple prediction tasks that boils down to guessing the probability of binary outcomes is that the QSR is not reliable in belief elicitations on complicated tasks with multiple outcomes (Palfrey and Wang [2009]).

17. Note that the default is blank, so even if they choose to sell nothing, they still need to click the dropdown menu to make a selection.

18. Weber and Welfens [2007] documented the disposition effect in an experiment with only selling decisions in a hypothetical real estate market.

19. The Claremont Colleges is a consortium of five undergraduate liberal arts colleges and two graduate institutions, situated in Southern California. The Institutional Review Board of the Claremont Graduate University approved the experimental procedures.

20. The detailed instructions and practice questions for the participants are available upon request.

21. This worked in the following way: the participant rolled a die, and if it came up with 1 or 2, the participant was paid the payoff associated with probability 1/3 in the lottery; if it came up with 3–6, the participant was paid the payoff associated with probability 2/3.

22. The SRM of the Predict condition is slightly positive for sequences ending in gains, but slightly negative for sequences ending in losses; and the opposite is true for the Both condition. This may suggest that the participants in the Predict condition were slightly more optimistic than in the Both condition. But the difference between the two conditions is not significant (p > 0.10). More studies are needed to test this further.

23. Specifically, the standard error is given by: , where , , , and are the number of realized gains, paper gains, realized losses, and paper losses (Odean [1998], p. 1784), respectively.

24. This was possibly due to the fact that the choice of the number of shares to sell was made by selecting from a drop down menu that started with 0 on top, so that participants might prefer to choose small numbers. However, even if they chose to sell nothing, they still needed to take some action, as the default was blank, which was not allowed. And the statistics on selling decisions (Table 2) indicate considerable variation, instead of a large amount of zeros.